DataArray transpose inconsistent with Dataset Ellipsis usage

[ahuang11]

This works:

import xarray as xr
ds = xr.tutorial.open_dataset('air_temperature')
ds.transpose('not_existing_dim', 'lat', 'lon', 'time', ...)

This doesn't (subset air):

import xarray as xr
ds = xr.tutorial.open_dataset('air_temperature')
ds['air'].transpose('not_existing_dim', 'lat', 'lon', 'time', ...)

The error message is a bit inaccurate too since I do have Ellipsis included; might be related to two calls of: dims = tuple(utils.infix_dims(dims, self.dims))

ValueError: ('not_existing_dim', 'lat', 'lon', 'time') must be a permuted list of ('time', 'lat', 'lon'), unless `...` is included

Traceback
...
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-5-793dfc1507ea> in <module>
      2 ds = xr.tutorial.open_dataset('air_temperature')
      3 ds.transpose('not_existing_dim', 'lat', 'lon', 'time', ...)
----> 4 ds['air'].transpose('not_existing_dim', 'lat', 'lon', 'time', ...)

~/anaconda3/envs/py3/lib/python3.7/site-packages/xarray/core/dataarray.py in transpose(self, transpose_coords, *dims)
   2035         if dims:
   2036             dims = tuple(utils.infix_dims(dims, self.dims))
-> 2037         variable = self.variable.transpose(*dims)
   2038         if transpose_coords:
   2039             coords: Dict[Hashable, Variable] = {}

~/anaconda3/envs/py3/lib/python3.7/site-packages/xarray/core/variable.py in transpose(self, *dims)
   1388         if len(dims) == 0:
   1389             dims = self.dims[::-1]
-> 1390         dims = tuple(infix_dims(dims, self.dims))
   1391         axes = self.get_axis_num(dims)
   1392         if len(dims) < 2 or dims == self.dims:

~/anaconda3/envs/py3/lib/python3.7/site-packages/xarray/core/utils.py in infix_dims(dims_supplied, dims_all)
    724         if set(dims_supplied) ^ set(dims_all):
    725             raise ValueError(
--> 726                 f"{dims_supplied} must be a permuted list of {dims_all}, unless `...` is included"
    727             )
    728         yield from dims_supplied

[dcherian]

ds.transpose('not_existing_dim', 'lat', 'lon', 'time', ...)

IMO this should raise an error too

[ahuang11]

ds.transpose('not_existing_dim', 'lat', 'lon', 'time', ...)

IMO this should raise an error too

I actually like it handling non_existing_dims automatically; maybe could be keyword though:
ds.transpose('not_existing_dim', 'lat', 'lon', 'time', ..., errors='ignore')

[dcherian]

Yes i think missing_dims="ignore" would be great. This would match the kwarg in isel (https://xarray.pydata.org/en/stable/generated/xarray.Dataset.isel.html)

[mesejo]

Hey! Can I work on this?

[ahuang11]

Sure.

…

On Sat, Jan 2, 2021, 6:38 AM Daniel Mesejo-León ***@***.***> wrote: Hey! Can I work on this? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#4647 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADU7FFVUS2TTAA6MRGX5G73SX4HSRANCNFSM4UMKLBWQ> .

[mesejo]

Hey, I was working on this and notice that transpose uses the method infix_dims to resolve the Ellipsis, should I put the expected behavior on infix_dims or only on transpose? The method infix_dims is also used in variable.transpose and in variable._stack_once. For me it seems right to put the new behavior on infix_dims to keep the behavior uniform, but I would like to know your opinion.

As a side-note I also noted that the return value of infix_dims is an iterator but on every usage is either converted to tuple or list, should I change the return value or keep it as it is?

[max-sixty]

Yes, agree this should be changed in infix_dims.

Fine to update the return type if it changes, but no need to coerce it premptively.

Thanks!